Regular inspection of Bridges is a necessary condition for bridge maintenance and safety. Traditional human detection is time-consuming and laborious. Using UAVs instead of manpower not only improves detection efficiency but also saves costs. Usually, underbridge environments are accompanied by poor GNSS signals, making reliable drone positioning under bridge challenging. The visual odometer, the classic drone positioning scheme, often fails in the ever-changing light noise. The ground-air system and structured light assisted localization algorithm designed in this paper can effectively improve the robustness of UAV positioning. We use ground visual beacon to provide positioning reference for UAV and introduce attention mechanism to improve the effect of feature extraction during beacon recognition. In addition, the establishment of structured light protection enhanced the stability of the system under weak light environment. Deep convolutional Neural Networks is used to identify candidate regions containing these markers. This region is projected onto an infrared image for positioning, and structured light ranging is employed for accurate beacon positioning. We also conduct modeling and error analysis to optimize the system's performance in spatial deployment. Real experiments and simulations are conducted under various conditions, demonstrating the system's stable performance in changing lighting environments and GNSS signal absence.